Journal of Heredity

Journal of Heredity Advance Access published March 3, 2010
! The American Genetic Association. 2010. All rights reserved.
For permissions, please email: [email protected].
Journal of Heredity
doi:10.1093/jhered/esq017
Experiments with Digital Organisms
on the Origin and Maintenance of Sex
in Changing Environments
DUSAN MISEVIC , CHARLES OFRIA,
AND
RICHARD E. LENSKI
Address correspondence to Dusan Misevic at the address above, or e-mail: [email protected].
Abstract
Many theories have been proposed to explain the evolution of sex, but the question remains unsettled owing to a paucity of
compelling empirical tests. The crux of the problem is to understand the prevalence of sexual reproduction in the natural
world, despite obvious costs relative to asexual reproduction. Here we perform experiments with digital organisms (evolving
computer programs) to test the hypothesis that sexual reproduction is advantageous in changing environments. We varied
the frequency and magnitude of environmental change, while the digital organisms could evolve their mode of reproduction
as well as the traits affecting their fitness (reproductive rate) under the various conditions. Sex became the dominant mode
of reproduction only when the environment changed rapidly and substantially (with particular functions changing from
maladaptive to adaptive and vice versa). Even under these conditions, it was easier to maintain sexual reproduction than for
sex to invade a formerly asexual population, although sometimes sex did invade and spread despite the obstacles to
becoming established. Several diverse properties of the ancestral genomes, including epistasis and modularity, had no effect
on the subsequent evolution of reproductive mode. Our study provides some limited support for the importance of
changing environments to the evolution of sex, while also reinforcing the difficulty of evolving and maintaining sexual
reproduction.
Key words: Avida, digital evolution, epistasis, evolution of sex, genetic architecture, modularity
Why sex? Sexual reproduction is costly and complicated,
yet it is widespread in the biological world (Bell 1982).
This paradox has long fascinated biologists and generated
a multitude of hypotheses and experimental tests
(Weismann 1889; Bell 1982; Michod and Levin 1988;
Kondrashov 1993; Rice 2002). Perhaps the simplest and
most intuitive explanation is that sex, by increasing
variation, can accelerate the rate of adaptation to novel or
changing environments (McPhee and Robertson 1970;
Malmberg 1977; Goddard et al. 2005; Cooper 2007).
Theoretical analyses have suggested that sex will be favored
especially when the fitness contributions of particular allele
combinations frequently switch between being advantageous and disadvantageous (Charlesworth 1976, 1993b;
Maynard Smith 1978). Such reversals in selection can be
caused by host–parasite interactions or other forms of
coevolution (Van Valen 1973; Hamilton 1980; Lively 1987;
Peters and Lively 1999; Lively and Dybdahl 2000; Salathe
et al. 2008). However, the more general theoretical
requirement is for changing environments, regardless of
the precise ecological basis (Sasaki and Iwasa 1987;
Charlesworth 1993a; Barton 1995; Kondrashov and
Yampolsky 1996; Otto and Michalakis 1998; Waxman and
Peck 1999; Otto and Nuismer 2004; Gandon and Otto
2007). We note that selection caused by parasites typically
acts in a frequency- or density-dependent manner, whereas
stochastic changes in the environment—even those that
reverse the direction of selection on particular traits—would
not necessarily act in the same way. Also, even if sexual
reproduction can accelerate adaptation to changing environments, such an effect might be more important in
maintaining sex than in facilitating its origin. That is, this
1
Downloaded from http://jhered.oxfordjournals.org at Peking University on August 19, 2010
From the Program in Ecology, Evolutionary Biology, and Behavior, Michigan State University, 103 Giltner Hall, East Lansing,
MI 48824 (Misevic, Ofria, and Lenski); the Department of Computer Science and Engineering, Michigan State University,
3115 Engineering Building, East Lansing, MI 48824 (Ofria); and the Department of Microbiology and Molecular Genetics,
Michigan State University, 2215 Biomedical Physical Sciences, East Lansing, MI 48824 (Lenski). Dusan Misevic is now at the
INSERM U1001, Centre de Recherches Interdisciplinaires (CRI), Faculté de Médecine, site Cochin Port-Royal, Université
Paris Descartes, 24 rue du Faubourg Saint Jacques, 75014 Paris, France.
Journal of Heredity
Methods
Experimental System
All experiments were conducted using Avida software
(available without cost at http://avida.devosoft.org/) with
the default settings, unless otherwise indicated. Digital
organisms in Avida are short self-replicating computer
programs that can reproduce either asexually or sexually,
depending on which divide instruction they execute. Digital
genomes are built from the default instruction set with 27
instructions including 2 divide instructions, divide-sex and
divide-asex, only one of which can be expressed by any
individual (Misevic et al. 2006). Mutations may affect the
organisms’ reproductive mode as well as their interactions
with substrates present in the environment. In this study,
point, insertion, and deletion mutations occurred at rates of
0.002, 0.0005, and 0.0005 per instruction copied, respectively, with the same mutation rates applied to the divide
instructions as all others; these are the default rates in Avida,
and they were used for consistency with previous studies
(Wilke et al. 2001; Ofria and Wilke 2004). The carrying
capacity (maximum population size) was 3600 organisms;
when a population was at carrying capacity, each new
offspring replaced a randomly chosen organism from
anywhere in the population. For the first 1000 updates of
each experimental run, the populations evolved in the same
constant environment with 9 substrates used previously
(Misevic et al. 2006) to evolve the organisms that served as
the ancestors in this study, after which additional and
changing substrates were introduced, as described below in
the section on Digital Metabolism. An update is a unit of time
in Avida during which an average of 30 instructions are
executed per organism in the population. A generation
typically requires 5–10 updates, with the precise number
depending on the genomic and phenotypic complexity of the
organisms in a population. Fitness was recorded for each
2
organism and then averaged over all organisms in the
population and log10 transformed for statistical analyses. All
of the experiments here started from digital organisms that
previously evolved in the experiments reported by Misevic
et al. (2006). We chose these populations because they are
well described and capture a broad range of genetic
properties, such as epistasis and modularity, that might
affect the evolution of sex. In addition, we chose the Avida
system for this study because, although computational in
nature, it is much more complex and biological than typical
numerical simulations. In essence, it provides an in silico
instantiation of open-ended evolution (Wilke and Adami
2002; Lenski et al. 2003; Ofria and Wilke 2004; Adami 2006).
Recombination Mechanism
After a digital organism has copied its genome (in whole or
in part, with or without mutations), the first execution of
either divide instruction determines its mode of reproduction and, simultaneously, separates the replicating genome
into 2 progeny genomes, as described previously (Misevic
et al. 2006). Progeny produced by the divide-sex instruction
undergo recombination, whereas those produced by divideasex are immediately placed at random locations in the
population. During sexual reproduction, 2 consecutively
produced progeny genomes are paired and exchange a single
continuous and corresponding region of their circular
genomes. These genomes are not necessarily aligned; the
identity of recombining regions is based only on the
correspondence in their genomic position and not on
sequence similarity or homology. The decision to employ
this mechanism was motivated by computational efficiency.
Recombination based on genomic position probably
increases the load of maladapted sexual offspring relative
to schemes based on sequence similarity or homology
because insertions and deletions disrupt the positional
correspondence of related genes. After the exchange of
corresponding genomic regions, both offspring are placed at
random locations in the population, in the same manner as
asexually produced organisms. In our study, organisms do
not experience the familiar 2-fold cost of sex because 2
sexual organisms produce 2 offspring. However, digital
organisms do experience certain intrinsic costs related to
recombination, such as disrupting favorable genetic combinations and combining incompatible genes.
Each incipient sexually produced offspring is stored until
another such offspring is produced, at which time they
recombine their genomes and the 2 recombinant products
are placed into the population. By default, the incipient
offspring can wait indefinitely, but supplementary experiments in which we limited the maximum waiting time did
not qualitatively change evolutionary outcomes (data not
shown). Therefore, we decided against introducing another
variable in this study.
Digital Metabolism
An organism’s genome may contain information that
encodes the ability to metabolize one or more substrates
Downloaded from http://jhered.oxfordjournals.org at Peking University on August 19, 2010
mechanism might provide an advantage to a sexual clade
relative to an asexual clade, but whether it can provide an
individual-level benefit for sexual reproduction is less clear.
Although a single asexual mutant might well be able to
invade a sexually reproducing population, 2 sexual organisms are required to gain any foothold in an asexual
population (absent the possibility of selfing). We are thus
interested in determining whether changing environments
can select for the maintenance as well as the initial spread of
sexual reproduction.
To explore the evolution of sexual reproduction in general,
and the role of changing environments in particular, one
would like an experimentally tractable evolving system with
varying environments and organisms that can mutate between
asexual and sexual forms. To that end, we have performed
experiments on populations of digital organisms in the Avida
system (Wilke and Adami 2002; Lenski et al. 2003; Ofria and
Wilke 2004; Adami 2006) to test whether environmental
change, including stochastically reversing the direction of
selection on some traits, promotes the evolution of sex.
Misevic et al. ! Sex In Changing Environments
successive generations. The particular pair of substrates that
changed their status at any point in time were chosen at
random (with a uniform probability distribution) from those
in each variable category.
Parameters Governing Environmental Change
We manipulated the extent of environmental change in 2
respects: by varying the frequency with which the substrates
changed from beneficial to harmful or vice versa (Figure 1)
and by varying the magnitude of the benefits and costs of
metabolizing the substrates. The intervals between changes
in the environment were 1, 3, 10, 30, 100, or 300 updates, and
the metabolic values between which substrates switched were
(–1, 1), (–1, 3), (–1, 5), or (–3, 1), where negative and positive
values indicate poisons and nutrients, respectively. For
example, when an environmental change occurs during a
(–1, 5) treatment, a random nutrient that we can call substrate
A is chosen and it becomes a poison, such that its metabolic
value changes from 5 to –1. Simultaneously, a random poison
that we can call substrate B is chosen and becomes a nutrient,
so that its value shifts from –1 to 5. Consider 4 organisms
Figure 1. Examples of 3 different periods of change in substrate metabolic values. The 9 fixed substrates and 68 changing
substrates are represented by different rows on the y axis. Black cells indicate that a substrate has a positive metabolic value
(nutrient), whereas white cells indicate that it has a negative value (poison) at that particular time in an experiment. The light gray
region at the left marks the first 1000 updates when all the changing substrates had a metabolic value of zero, allowing the
populations to acclimate to their ancestral condition in which only the 9 fixed nutrients were present. Panels (a), (b), and (c)
represent environmental change at periods of 3, 30, and 300 updates, respectively.
3
Downloaded from http://jhered.oxfordjournals.org at Peking University on August 19, 2010
present in its environment (Wilke and Adami 2002; Lenski
et al. 2003; Ofria and Wilke 2004). Metabolism of a substrate
either accelerates or decelerates an organism’s replication
rate by a factor of 2m, where m is the substrate’s metabolic
value and is positive or negative for a nutrient or poison,
respectively. Periodically, one randomly chosen substrate
switches from nutrient to poison whereas another substrate
simultaneously undergoes the opposite transition. A total of
77 substrates are associated with all of the distinct 1-, 2-, and
3-input logic operations. An organism metabolizes one of
these substrates when it executes a sequence of genomic
instructions that perform the associated logic operation.
Nine substrates (representing the 1- and 2-input operations)
that were present during the prior evolution of the ancestors
used in our study were again present and always nutritious.
The other 68 substrates alternated randomly between being
nutrients and poisons, with 25 being nutritious at each
moment, according to the treatment schedules described in
the next section below. Although some treatment schedules
involved changes at periods shorter than generation times,
each change affected only 2 of the 68 variable substrates, so
that most aspects of the environment were constant across
Journal of Heredity
that have different genotypes but identical phenotypes,
except for the following properties: genotype GØ metabolizes neither substrate A nor substrate B, GA metabolizes
only A, GB metabolizes only B, and GAB metabolizes both A
and B. Table 1 shows how the fitness of each of these
organisms would be affected by this environmental change.
For each of the 24 experimental treatments (6 rates of
change, 4 magnitudes of change), we propagated 20 initially
sexual and 20 initially asexual populations for 100 000
updates, giving a total of 960 populations.
Properties of Ancestral Organisms
Table 1 Illustration of the effects of an environmental change
on the fitness of 4 different genotypes under the (–1, 5) valuation
scheme for poisons and nutrients, respectively
GØ fitness GA fitness GB fitness GAB fitness
Before change W0
After change W0
W0 # 25 W0 # 2$1 W0 # 24
W0 # 2$1 W0 # 25 W0 # 24
The 4 types are phenotypically identical except for their abilities to
metabolize substrates A and B. Genotype GØ is unable to metabolize either
A or B, and it has fitness W0. By contrast, GA metabolizes only A, GB
metabolizes only B, and GAB metabolizes both A and B. Values indicate
each genotype’s fitness before (first row) and after (second row) an
environmental change whereby substrate A turns from nutrient into poison
and, simultaneously, substrate B turns from poison into nutrient.
4
Results and Discussion
We begin by examining the rates and magnitudes of
environmental change that favor the evolution of sexual
reproduction. We then compare populations that started
with different modes of reproduction to distinguish between
the origin and maintenance of sex. We also analyze the
possible influence of genetic architecture on the propensity
to evolve sex. Finally, we compare the fitness effects of
sexual and asexual reproduction.
Effects of Changing Environment on Reproductive Mode
The trajectories for the relative abundance of sexual and
asexual organisms were highly variable in our experiments,
even among replicate populations in the same treatment
(Figure 2). Some populations began and mostly remained
either sexual (Figure 2h) or asexual (Figure 2c), whereas
others switched their mode of reproduction multiple times
(e.g., Figure 2f). Sexual and asexual types typically coexisted
only for a short period during a transition between
reproductive modes, but occasionally both types maintained
intermediate frequencies for much longer periods
(Figure 2b).
Figure 3a summarizes the final numerically dominant
mode of reproduction, averaged more than the 40 replicate
populations (with half initially sexual and half initially
asexual) for each of the 24 treatments. In this analysis, each
population was classified as either sexual or asexual based
on the most abundant mode of reproduction at the end of
the experiment. Asexual reproduction prevailed under most
of the conditions tested. However, sexual reproduction
tended to be relatively more common at faster rates of
environmental change (i.e., as the period of change was
shorter), provided also that the benefit of substrates when
they were nutrients was large and their cost when poisonous
was not. In the most extreme case, when the period of
change was 1 update, and the metabolic values of poisons
and nutrients were –1 and þ5, respectively, 65% of the
populations were predominantly sexual at the end of the
experiment.
As noted earlier, some populations repeatedly switched
their reproductive mode during the evolution experiment
(Figure 2). To avoid a potential artifact resulting from the
arbitrary choice of the experiment’s duration, we also
calculated the proportion of time during which the majority
of a population was reproducing sexually. Thus, a population
that spent most of its history as asexual would be classified
as such, even if it became sexual shortly before the end of
the experiment (e.g., Figure 2d). The overall trends are
similar, however, whether one considers the final state
(Figure 3a) or the proportion of time a population spent
Downloaded from http://jhered.oxfordjournals.org at Peking University on August 19, 2010
Our work builds on the experiments of Misevic et al. (2006),
in particular information on the genetic architecture of the
digital organisms used to seed the populations in this study.
Therefore, we must briefly describe our previous results
here. The populations in our earlier work evolved either
strictly sexually or strictly asexually for more than 10 000
generations. For each evolved population, we measured its
average fitness, genome length, modularity, fitness effects
of, and interactions between random mutations. We showed
that sexual genomes tended to be longer, on average, and
had evolved greater modularity in 2 respects. First, those
regions of the genome responsible for metabolizing
particular nutrients were typically more compact (higher
physical modularity, a measure of the mean distance
between sites encoding computational traits) in sexual than
in asexual organisms. Second, those regions encoding
distinct metabolic functions had less overlap (higher
functional modularity, a measure of the average overlap in
the genomic sites encoding different traits) in sexual than in
asexual organisms. Precise definitions of physical and
functional modularity can be found in Misevic et al.
(2006). To quantify the effects of random mutations as
well as their interactions, we constructed millions of single
and multiple mutants in each evolved background and
measured their fitness values. We fit a power function to
those data, log10W 5 –aMb, where W is the average fitness
and M is the corresponding number of random mutations.
Parameters a and b reflect robustness to individual
mutations and the form of mutational interactions (epistasis), respectively. We found that the sexually evolved
organisms were, on average, more robust to mutations
(asex , aasex) than were the asexual organisms. Interactions
among mutations were, on average, antagonistic in both
sets, but less so in the sexual than in the asexual organisms
(basex , bsex , 1).
Misevic et al. ! Sex In Changing Environments
dominated by sexually reproducing organisms (Figure 3b).
Using the latter data set (in which the dependent variable is
continuous rather than discrete), the effects of the period
and magnitude of environmental change are both highly
significant (2-way analysis of variance [ANOVA]: F5,936 5
17.417, P , 0.001 for the period; F3,936 5 6.135,
P , 0.001 for the magnitude), as is their interaction
(F15,936 5 10.502, P , 0.001).
Origin versus Maintenance of Sex
Many authors have reasoned that it is easier to maintain
sexual reproduction than for it to evolve de novo given the
costs of sexual reproduction (Michod and Levin 1988;
Lenski 1999). By starting our experiments with populations
of digital organisms that were either entirely sexual or
entirely asexual, we can compare the conditions that allow
the origin versus maintenance of sex. As noted in the
Methods, these digital organisms do not face the widely
discussed 2-fold cost of sex because the 2 parental genomes
recombine to produce 2 offspring that are both placed in
the population. However, selfing cannot occur in this
system, and thus sexual reproduction requires 2 parents
who, moreover, must be able to produce viable recombinant
offspring, perhaps making it difficult for sex to gain
a foothold in an asexual population. By contrast, only one
asexual mutant is needed to invade a sexual population.
Thus, one might expect an asymmetry between the origin
and maintenance of sex in this system, even without the
2-fold cost of sex.
To examine this issue, we increased the number of
replicates to 50 initially sexual and 50 initially asexual
populations for the subset of treatments with changing
environments where sex was, on average, most successful,
specifically with metabolic values switching between –1
and þ5 and with all 6 rates of change used previously, for
a total of 600 populations. Across all rates of environmental
change, populations were more likely to be predominantly
sexual at the end of the experiment if they were initially
sexual than if they started as asexual (Figure 4a). Sexual
organisms also dominated the populations that started as
sexual for a greater proportion of the total time than they
did in populations that began as asexual (Figure 4b). The
effect of the initial state on the time-averaged dominant
mode of reproduction during evolution in changing
environments was highly significant (2-way ANOVA,
F1,588 5 32.160, P , 0.001), whereas the interaction
between the initial mode of reproduction and the rate of
environmental change was not significant (F5,588 5 0.850,
P 5 0.514). Sex was also more common, regardless of
ancestral mode, in more rapidly changing environments
(F5,588 5 29.237, P , 0.001).
5
Downloaded from http://jhered.oxfordjournals.org at Peking University on August 19, 2010
Figure 2. Heterogeneity in the proportion of sexual organisms over time in replicate evolving populations. Ancestors were
asexual (blue) for panels (a–d) and sexual (red) for panels (e–h). All 8 populations evolved under the same 3-update period and
(–1, 5) magnitude of costs and benefits of metabolizing the 68 changing substrates.
Journal of Heredity
The effect of initial reproductive state might simply
indicate a lag time required for the origin and spread of the
alternative mode. Alternatively, the evolutionary history
prior to the ancestral state might have produced genomic
constraints that make it more difficult to switch reproductive mode while retaining overall fitness. To address this
issue, we reanalyzed the data by calculating the proportion
of time that populations were predominantly sexual during
only the second half of the experiments. We observed the
same qualitative pattern, with significant effects of the initial
mode of reproduction and period of change, but no
significant interaction between them (data not shown),
though the effect of the initial mode of reproduction was
somewhat reduced. Over the entire duration of the
experiment, the populations with sexual ancestors were
6
Figure 4. Relationship between the final prevalence of sex
and the ancestral mode of reproduction. All populations
evolved in the environment with (–1, 5) magnitude of costs and
benefits, while the period of environmental change is shown
along the x axis. For each period, the final proportion of
predominantly sexual populations (panel 4a) and proportion of
time spent sexual (panel 4b) were measured in 50 populations
that were initially sexual (squares) and 50 others that were
initially asexual (diamonds). Error bars show one standard error
of the mean.
predominantly sexual 38% more often than those with
asexual ancestors; in the second half of the experiment, this
difference was reduced to 25%. Thus, both factors seem to
contribute to the effect of initial state on the evolution of
reproductive mode. In any case, sexual reproduction
overcame the barriers that hindered its establishment in
previously asexual populations about half the time in the
most favorable treatments (Figure 4).
Effects of Genetic Architecture on the Evolution of Sex
Other studies have shown that the reproductive mode
influences the evolution of various aspects of genetic
architecture including robustness, epistasis, and modularity
Downloaded from http://jhered.oxfordjournals.org at Peking University on August 19, 2010
Figure 3. Prevalence of sexual reproduction for populations
that evolved in environments with different periods and
magnitudes of environmental change. (a) Populations were
categorized as sexual if more than half of the organisms at the
end of the experiment reproduced sexually. (b) Average
proportion of time that evolving populations spent with
predominantly sexual reproduction. Error bars represent one
standard error of the mean.
Misevic et al. ! Sex In Changing Environments
Fitness of Sexual and Asexual Populations
We also sought to determine whether sexual organisms
achieved higher fitness than asexual ones, especially under
those conditions where sexual reproduction was most
successful. To do so, we compared the average fitness
Table 2 Discriminant analyses for the final mode of
reproduction under 6 different periods of environmental change
Period of
change
Wilks’s lambda
F7,92
P
% Correct
1 update
3 updates
10 updates
30 updates
100 updates
300 updates
0.983
0.929
0.902
0.880
0.945
0.981
0.267
1.186
1.694
2.116
0.911
0.298
0.951
0.321
0.131
0.059
0.491
0.937
38
57
56
63
44
47
Each discriminant function was constructed using the ancestor’s fitness,
genome length, physical modularity, functional modularity, robustness to
individual mutations (a), and coefficient of epistasis (b). P values indicate
whether the overall function was significant. Percent correct shows the
proportion of 100 evolved populations that were correctly classified as
predominantly sexual or asexual at the end of the experiment using the
discriminant function.
values of sexual and asexual populations that evolved with
the (–1, 5) magnitude of change under the 6 different rates
of change. We identified the periods of time for each
population during which the majority of organisms were
either sexual or asexual, and we calculated the mean fitness
values during those periods (Figure 5). (Averaging fitness
over time is reasonable because, although the environments
frequently changed, the numbers of nutrients and poisons
did not change.) The mode of reproduction and period of
Figure 5. Average fitness of predominantly sexual and
predominantly asexual populations. The log10-transformed
fitness values while a particular population was predominantly
sexual or asexual were separately averaged. The time-averaged
fitness levels while sexual (squares) or asexual (diamonds)
were then averaged more than the 100 populations that
evolved at each period of environmental change. All
populations evolved in the environments with (–1, 5)
magnitude of costs and benefits of metabolizing the
68 changing substrates. Error bars show one standard error of
the mean.
7
Downloaded from http://jhered.oxfordjournals.org at Peking University on August 19, 2010
(Azevedo et al. 2006; Misevic et al. 2006; Lohaus et al. 2010).
Here we ask, conversely, whether these features, as reflected
in a particular ancestral state, might influence a population’s
propensity to evolve sexual reproduction. As shown in the
section above, the evolution of sexual reproduction is more
difficult than its maintenance. Moreover, organisms may
have evolved genetic architectures that are coadapted to
their reproductive mode, thus making it more difficult to
switch to the alternative mode. For example, favorable gene
combinations assembled in asexual lineages might be
disrupted by recombination, reducing the fitness of
recombinant offspring, thereby impeding the evolution of
sex. On the other hand, certain genetic architectures might
promote sexual reproduction. First, more modular genomes
may allow faster exchange of metabolic building blocks,
making sexual organisms more evolvable (Wagner and
Altenberg 1996; Earl and Deem 2004; Schlosser and Wagner
2004; Sun and Deem 2009). Second, synergistic epistatic
interactions between deleterious mutations may facilitate the
removal of those mutations through recombination and
selection, thus promoting sexual reproduction (Kondrashov
1982; Michod and Levin 1988; De Visser and Hoekstra
1998; Wolf et al. 2002).
To determine whether modularity, epistasis, or other
properties of the ancestors (besides reproductive mode
itself) influenced the evolution of sex in our experiments, we
further analyzed the 600 populations that evolved with the
(–1, 5) magnitude of change under the 6 different periods of
environmental change. In particular, we ran discriminant
analyses separately for each of these 6 treatments, where the
objective was to obtain functions that would categorize the
final populations as either sexual or asexual based on
ancestral fitness, genome length, physical modularity,
functional modularity, robustness to individual mutations
(a), and strength of epistasis (b). (The measurements of
these ancestral properties were performed by Misevic et al.
2006 on 50 sexual and 50 asexual populations that evolved
independently for 100 000 updates; see methods and
Misevic et al. 2006 for more detailed descriptions of these
metrics and previous experiments.) For these analyses, the
final population was categorized as sexual or asexual based
on the most common reproductive mode at the end of the
experiment. The 6 discriminant functions, on average,
correctly classified only 50.83% of the populations—hardly
better than random guesses—and none was significant
(Table 2), even without adjusting significance levels for
multiple tests of the same hypothesis. In agreement with
recent theoretical work (Misevic et al. 2009), our results
therefore indicate that genetic architecture is not strongly
predictive of an evolving population’s eventual reproductive
mode.
Journal of Heredity
Conclusions
Our experiments show that rapidly changing environments
can promote the evolution of sex, at least relative to more
slowly changing environments, in this artificial system. At
the same time, our results call attention to several limitations
of the theory that changing environments will favor the
evolution of sexual reproduction: the parameter space that
favors sex is quite limited (Figure 3); the origin of sexual
reproduction is more difficult than its maintenance (Figure 4);
and idiosyncratic effects of ancestry and chance exert
strong influences on whether sex evolves (Figure 2). The
same or similar limitations are relevant for other theories for
the evolution of sex (Otto and Feldman 1997; Otto and
Nuismer 2004; Misevic et al. 2004), which has led some
researchers to conclude that multiple factors are necessary
to account for the evolution of sex (West et al. 1999). In any
case, our study shows the utility of digital organisms for
testing complex evolutionary theories because they allow
one to manipulate relevant features of the environment,
control for confounding effects of ancestry, compare the
origin and maintenance of organismal traits under the same
conditions, and obtain data across many replicate populations for thousands of generations. Of course, any insights
gained from experiments with digital organisms may also
suggest more focused research on biological systems to
examine the generality of those insights.
References
Adami C. 2006. Digital genetics: unravelling the genetic basis of evolution.
Nat Rev Genet. 7:109–118.
Azevedo RBR, Lohaus R, Srinivasan S, Dang KK, Burch CL. 2006. Sexual
reproduction selects for robustness and negative epistasis in artificial gene
networks. Nature. 440:87–90.
Barton NH. 1995. A general model for the evolution of recombination.
Genet Res. 65:123–144.
Bell G. 1982. The masterpiece of nature. Berkeley (CA): University
California Press.
Charlesworth B. 1976. Recombination modification in a fluctuating
environment. Genetics. 83:181–195.
Charlesworth B. 1993a. Directional selection and the evolution of sex and
recombination. Genet Res. 61:205–224.
Charlesworth B. 1993b. The evolution of sex and recombination in
a varying environment. J Hered. 84:345–350.
Cooper TF. 2007. Recombination speeds adaptation by reducing
competition between beneficial mutations in populations of Escherichia coli.
PLoS Biol. 5:1899–1905.
De Visser J, Hoekstra RF. 1998. Synergistic epistasis between loci affecting
fitness: evidence in plants and fungi. Genet Res. 71:39–49.
Earl DJ, Deem MW. 2004. Evolvability is a selectable trait. Proc Natl Acad
Sci U S A. 101:11531–11536.
Gandon S, Otto SP. 2007. The evolution of sex and recombination in
response to abiotic or coevolutionary fluctuations in epistasis. Genetics.
175:1835–1853.
Goddard MR, Charles H, Godfray J, Burt A. 2005. Sex increases the
efficacy of natural selection in experimental yeast populations. Nature.
434:636–640.
Hamilton WD. 1980. Sex versus non-sex versus parasite. Oikos.
35:282–290.
Kondrashov AS. 1982. Selection against harmful mutations in large sexual
and asexual populations. Genet Res. 40:325–332.
Kondrashov AS. 1993. Classification of hypotheses on the advantage of
amphimixis. J Hered. 84:372–387.
Kondrashov AS, Yampolsky LY. 1996. Evolution of amphimixis and
recombination under fluctuating selection in one and many traits. Genet
Res. 68:165–173.
Lenski RE. 1999. A distinction between the origin and maintenance of sex.
J Evol Biol. 12:1034–1035.
Lenski RE, Ofria C, Pennock RT, Adami C. 2003. The evolutionary origin
of complex features. Nature. 423:139–144.
Lively CM. 1987. Evidence from a New Zealand snail for the maintenance
of sex by parasitism. Nature. 328:519–521.
Lively CM, Dybdahl MF. 2000. Parasites adaptation to locally common host
genotypes. Nature. 405:679–681.
Funding
US National Science Foundation (CCF-0643952) to C.O.;
DARPA ‘Fun Bio’ Program (HR0011-05-1-0057) to R.E.L.
and C.O.
Acknowledgements
We thank Ricardo Azevedo and 2 anonymous reviewers for their helpful
comments on an earlier version of our paper. D.M. also thanks John
Logsdon and Maurine Neiman for organizing the symposium at which this
work was presented.
8
Lohaus R, Burch CL, Azevedo RBR. 2010. Genetic architecture and the
evolution of sex. J Hered.
Malmberg RL. 1977. Evolution of epistasis and advantage of recombination
in populations of bacteriophage T4. Genetics. 86:607–621.
Maynard Smith J. 1978. The evolution of sex. Cambridge (UK): Cambridge
University Press.
McPhee CP, Robertson A. 1970. The effect of suppressing crossing-over on
the response to selection in Drosophila melanogaster. Genet Res. 16:1–16.
Michod RE, Levin B, editors. 1988. The evolution of sex. Sunderland (MA:
Sinauer.
Misevic D, Kouyos RD, Bonhoeffer S. 2009. Predicting evolution of sex on
complex fitness landscapes. PLoS Comp Biol. 5:e1000510.
Downloaded from http://jhered.oxfordjournals.org at Peking University on August 19, 2010
environmental change had significant effects on fitness, as
did their interaction (2-way ANOVA: F1,1188 5 8.930,
P 5 0.003 for reproductive mode; F5,1188 5 8.068, P , 0.001
for period of change; F5,1188 5 2.882, P 5 0.014 for
interaction). In the rapidly changing environments, those
populations dominated by sexual organisms typically had
higher average fitness than did predominantly asexual
populations (Figure 5). By contrast, in the slowly changing
environments, although most populations evolved to be
predominantly asexual (Figure 3), the individuals in sexual
and asexual populations achieved similar average fitness
levels (Figure 5). This pattern calls attention to potential
discrepancies between individual and population metrics of
success when it comes to reproductive mode, an issue that
merits further study.
Misevic et al. ! Sex In Changing Environments
Misevic D, Ofria C, Lenski RE. 2004. Sexual reproduction and Muller’s
ratchet in digital organisms. In Proceedings of Artificial Life IX. In: Pollack
JB, Bedau M, Husbands P, Ikegami T, Watson RA, editors. Cambridge
(MA): MIT Press. pp. 340–345.
Schlosser G, Wagner GP, editors. 2004. Modularity in development and
evolution. Chicago (IL): University Chicago Press.
Misevic D, Ofria C, Lenski RE. 2006. Sexual reproduction reshapes the
genetic architecture of digital organisms. Proc R Soc Lond B Biol Sci.
273:457–464.
Van Valen L. 1973. A new evolutionary law. Evol Theor. 1:1–30.
Ofria C, Wilke CO. 2004. Avida: a software platform for research in
computational evolutionary biology. Artif Life. 10:191–229.
Otto SP, Felman MW. 1997. Deleterious mutations, variable epistatic
interactions, and the evolution of recombination. Theor Popul Biol.
51:134–147.
Otto SP, Nuismer SL. 2004. Species interactions and the evolution of sex.
Science. 304:1018–1020.
Peters AD, Lively CM. 1999. The Red Queen and fluctuating epistasis:
a population genetic analysis of antagonistic coevolution. Am Nat.
154:392–405.
Rice WR. 2002. Experimental tests of the adaptive significance of sexual
recombination. Nat Rev Genet. 3:241–251.
Wagner GP, Altenberg L. 1996. Complex adaptations and the evolution of
evolvability. Evolution. 50:967–976.
Waxman D, Peck JR. 1999. Sex and adaptation in a changing environment.
Genetics. 153:1041–1053.
Weismann A. 1889. Essays upon heredity and kindred biological problems.
Oxford: Clarendon Press.
West SA, Lively CM, Read AF. 1999. A pluralist approach to sex and
recombination. J Evol Biol. 12:1003–1012.
Wilke CO, Adami C. 2002. The biology of digital organisms. Trends Ecol
Evol. 17:528–532.
Wilke CO, Wang JL, Ofria C, Lenski RE, Adami C. 2001. Evolution of
digital organisms at high mutation rates leads to survival of the flattest.
Nature. 412:331–333.
Wolf JB, Brodie ED, Wade MJ, editors. 2002. Epistasis and the
evolutionary process. Oxford: Oxford University Press.
Salathe M, Kouyos RD, Bonhoeffer S. 2008. The state of affairs in the
kingdom of the Red Queen. Trends Ecol Evol. 23:439–445.
Received October 8, 2009; Revised January 14, 2010;
Accepted January 27, 2010
Sasaki A, Iwasa Y. 1987. Optimal recombination rate in fluctuating
environments. Genetics. 115:377–388.
Corresponding Editor: John Logsdon
9
Downloaded from http://jhered.oxfordjournals.org at Peking University on August 19, 2010
Otto SP, Michalakis Y. 1998. The evolution of recombination in changing
environments. Trends Ecol Evol. 13:145–151.
Sun J, Deem MW. 2009. Spontaneous emergence of modularity in a model
of evolving individuals. Phys Rev. E 79:031907.